A Brief History of Netflix Personalization

From startup in 1998 to today, a detailed history of the strategy, metrics, and experiments Netflix executes to develop a personalized experience focused on delivering its members movies they love

Photo: freestocks via Unsplash

Introduction

1998: A DVD-by-mail startup is born

Netflix in 1999 had 2,600 DVDs to choose from. The service was a la carte. Netflix didn’t become an “all you can eat” subscription service until 2000.

2000: Cinematch

2001: The five-star rating system

The red stars indicate the expected rating for a member, although most assumed It was the average rating.

2002: Multiple algorithms

2004: Profiles

2004: Netflix launches “Friends”

Friends let you connect with friends on the service, then see what they watched and what they recommended to you.

2006: The Netflix personalization strategy

The early proxy metric for Netflix’s personalization efforts was the percentage of members who rated at least 50 movies or TV shows during their first two months with the service.

2006: The ratings wizard

2006: Demographic data

Root Mean Square Error is a calculation of the predicted versus actual ratings for all Netflix members. It’s a “down and to the right” metric that got better over time.

2006: Collaborative filtering in the QUACL

The QUACL, or Queue Add Confirmation Layer, was a significant source of personalized merchandising.

2006: The $1 million Netflix Prize

Netflix offered $1 million to any team that could improve RMSE by at least 10%.

2007: Netflix streaming launch

Netflix launched streaming in 2007. It was a free add-on for the DVD-by-mail service. The first streaming-only offering launched in Canada in 2010.

2007: Netflix Prize

The leaderboard for the Netflix Prize at its conclusion in 2009. The fourth-place team, “Opera Solutions and Vandelay United,” was the combined submission of three teams and included an insider joke from “Seinfeld.”

2009: The next big Netflix prize

2010: Testing the new “Netflix Prize” algorithm

2010: Popularity matters

2011: Netflix’s movie genome project

Early execution of the “Category Interest” algorithm, circa 2011.

2011: How the personalization algorithms work

2011: Netflix proves personalization improved retention

2012: Profiles reinvented

2013: “House of Cards” original content launch

“House of Cards,” Netflix’s first big investment in original content, was actually an American remake of the British TV Series.

2013: Netflix wins a technical Emmy

2015: Does it matter if you’re French?

A Netflix PC-based homepage, en francais.

2016: Netflix tests a personalized interface

Note that row names that Netflix presents to you are wildly inconsistent. In this screen capture for me on PC, Netflix leads with a row entitled “Trending Now” and puts the “Continue Watching” row in the second position. The interface changes each time you use it, depending on your recent activity, platform, and time of day.

2017: From stars to thumbs

In 2017, Netflix tested its five-star ratings system against Facebook’s thumbs up/down interface. “Thumbs” doubled the amount of taste input from members.

2017: What happens to the five-star system?

This was one of the movies in Adam Sandler’s $250 million-dollar, four-picture original content deal. Sometimes movie enjoyment is all about leaving your brains at the door with a good three-star movie.

2017: Percentage match

Netflix no longer indicates the quality of a movie with its five-star system. It simply indicates the extent to which a title is a match for you, giving an indication of how much they think you will enjoy it.

2018: Personalized movie art

If you’re into romantic comedies, Netflix uses the top-right movie art when it merchandises “Good Will Hunting” to you. If you like famous comedians, your personalized movie art features Robin Williams.

2021: Do you feel lucky?

Conclusion

One last thing!

Other essays I think you’ll enjoy

Former VP/CPO at Netflix/Chegg. Now speaker, teacher, & workshop host. Learn more here: www.gibsonbiddle.com or here: https://www.linkedin.com/in/gibsonbiddle/